ACF: An Armed CCTV Footage Dataset for Enhancing Weapon Detection

Area of Science:

• Computer Vision
• Artificial Intelligence
• Surveillance Technology

Background:

• Rising crime rates necessitate advanced security measures.
• Closed-circuit television (CCTV) is crucial for public safety surveillance.
• Detecting small weapons in CCTV footage presents significant challenges due to object size.

Purpose of the Study:

• To develop an effective deep learning method for detecting small weapons in CCTV images.
• To address the limitations of existing public datasets for weapon detection.
• To introduce a novel image tiling technique for enhanced object detection.

Main Methods:

• Collection of a custom Armed CCTV Footage (ACF) dataset with mock pedestrian weapon scenarios.
• Implementation of an image tiling-based deep learning approach for small weapon object detection.
• Evaluation of the proposed method on a public benchmark dataset (Mock Attack) and using SSD MobileNet V2.

Main Results:

• The image tiling approach significantly improved detection performance, achieving a 10.22 times better mean Average Precision (mAP).
• On the ACF Dataset using SSD MobileNet V2, the method achieved an mAP of 0.758 for pistol and knife detection.
• The proposed method demonstrated a substantial enhancement in detecting small weapon objects.

Conclusions:

• The image tiling technique combined with the ACF dataset effectively enhances small weapon detection in CCTV footage.
• This approach offers a promising solution for improving the accuracy and efficiency of surveillance systems.
• The study highlights the potential of tailored datasets and innovative deep learning methods for security applications.